Eliminating child mortality remains one of the largest challenges facing the world. This issue disproportionately affects people and children living in low and middle-income countries. Access to clean drinking water is thought to be a major factor in reducing child mortality, but is this necessary the case?
Child mortality and drinking water access will be negatively correlated.
Other factors will also be statistically significant to child mortality.
Demographic Health Survey (DHS) data was downloaded from the United States Agency for International Development’s (USAID) website for all countries and years available. Key variables included child mortality rate, access to improved drinking water sources, access to improved sanitation, access to permanent hand washing systems, wealth quartiles, and household sizes. 358 observations are available for 87 different low and middle-income countries over the period of 1985-2022. Child mortality is expressed as a rate of deaths of children under the age of 5 years per 1,000 live births. Access to improved drinking water sources, sanitation, and permanent handwashing systems are expressed as a percentage of the population who have access to them. Wealth quartiles are available as the percentage of people who fall into the lowest, middle, and highest wealth quartiles in each country. All data were collected through health surveys. Child mortality ranges from 318 (Niger, 1992) to 6 (Armenia, 2015) and water access ranges from 16.9% (Burkina Faso, 1993) to 99.9% (Eygpt, 2005, 2008; Armenia, 2015).
Data was cleaned in Excel and imported into R Studio for analysis. Appropriate visuals were created and linear regression was performed to assess the two expectations. The linear regression contains three models: Model (1) tests the correlation between child mortality and water access, Model (2) adds sanitation access as a control to Model (1), and Model (3) adds lowest wealth quartile percentages and household size as controls to the previous models. Country fixed effects were added to the models in order to account for changes between individual country values between years. Both ggplot 2 and patchwork packages were utilized to create the figures.
Based upon the available data, child mortality is currently declining. Fig. 1 provides child mortality rates over time.
Figure 1: Child Mortality Over Time
There appears to be sharp contrast between overall water access, urban water access, and rural water access. Fig. 2 compares child mortality with water access stratified by urban and rural classifications.
Figure 2: Child Mortality Compared to Improved Drinking Water Source Access (%)
Table 1. provides the regression table produced in the analysis. It includes three models controlling for different variables.
| (1) | (2) | (3) | |
|---|---|---|---|
| * p < 0.05 | |||
| Water Access | -2.139* | -1.574* | -1.348* |
| Sanitation Access | -0.581* | -0.678* | |
| Lowest Wealth Qu. (Urban) | -0.376 | ||
| Lowest Wealth Qu. (Rural) | 0.198 | ||
| Household Size | 2.840 | ||
| Num.Obs. | 283 | 279 | 271 |
According to the linear regression model, drinking water access is negatively correlated with child mortality by a factor of 2.139. However, water access’ effect on child mortality declines when controlled for improved sanitation system access to -1.574. Interestingly, the percentage of people in both urban and rural areas in the lowest wealth quartile were insignificant to child mortality. This result is surprising. Additionally, when controlled for average household size, only water and sanitation access remains significant out of all the regression variables. Based upon these results, water and sanitation access are the only significant factors affecting child mortality rates.
Yes, child mortality and drinking water access were negatively correlated, even after controlling for various other variables.
Yes and No, sanitation access was the only other factor to be statistically significant to child mortality. I was expecting other variables, such as the wealth quartiles, to be significant, but they were not.